Support for Automating Speaker Labeling

Our transcription server uses Deepgram for transcribing audio, and it performs well with diarization. However, it does not support speaker profiles (or equivalent functionality). This means that while Deepgram identifies different speaker segments (e.g., "Speaker 0," "Speaker 1"), we must manually listen to the audio and replace these labels with the actual speaker names.

For most transcripts, we already know the speakers beforehand. Since some speakers appear in multiple transcripts, having a way to leverage speaker profiles (or another mechanism for identifying speakers) would allow us to automate this process and reduce manual effort.

Currently, manual review ensures accuracy, so this limitation is not critical. However, if we move away from human review in the future, automating speaker labeling would become essential to maintain efficiency and accuracy.

**Possible Solutions**
Deepgram does not prioritize this feature based on [this discussion](https://github.com/orgs/deepgram/discussions/534). As a potential alternative, we could investigate integrating [PyAnnote](https://github.com/pyannote) or similar solutions to address this gap.